In the 2026 short video landscape, AI-generated short dramas have already consumed over ninety percent of new releases, and a technology that lets creators "one person, one machine" produce dramas is attracting real money from listed companies.

On the evening of September 29, Wangsu Science & Technology Co.,Ltd. (300017.SZ) released an external investment announcement. This seventeen-year veteran of CDN (Content Delivery Network) announced plans to use 300 million yuan of its own funds to subscribe to A+ round preferred shares newly issued by video generation model company Sand.ai. After the transaction is completed, it will hold a 4.3963% stake.

The target is no small player. Founder Cao Yue is a Tsinghua Special Scholarship winner, co-first author of Swin Transformer, and an ICCV Marr Prize recipient. The company just open-sourced the world's first hundred-billion-parameter MoE video generation model in August.

On one side is an infrastructure provider whose growth has peaked and urgently needs a new story. On the other is a technology upstart with halo and credentials, struggling to survive in the cracks between giants. When Wangsu Science & Technology's main business growth has peaked, how does this veteran infrastructure provider use its smallest chip to bet on the most crowded and cash-burning track?

As of October 9, Wangsu Science & Technology Co.,Ltd. (300017.SZ) closed at 14.23 yuan per share, up 2.82%, with a total market value of 35.514 billion yuan.

Testing the Waters of a New Track with 300 Million Yuan

Those AI short dramas you scroll past on your phone are quietly becoming a big business. Bona Film Group teamed up with Douyin to launch "Sanxingdui: Future Revelation," using Jimeng AI to turn ancient Shu civilization into a sci-fi micro short drama, with Douyin views surpassing 135 million. Kuaishou's "Shanhai Qijing: Cleaving Waves" was generated by Kling AI, exceeding 50 million views in two weeks. CCTV Video's "Chinese Mythology" had its entire process from art, storyboarding to voiceover and music completed by AI, with nearly 20 million views across the web.

These three viral dramas came from peers like Jimeng and Kling, proving that video generation can already produce hits. But Sand.ai takes a different, lighter path. It turned the technology into VidMuse, accessible even to ordinary people. You just throw a song into it, and within minutes you get a music MV with storyboards and visuals.

The math on Wangsu Science & Technology's investment is very clear. According to the September 29 announcement, 300 million yuan converts to 44.4517 million USD, corresponding to 21.1137 USD per share, subscribing to 2.1054 million shares. Based on this investment ratio, Sand.ai's post-investment valuation for this round is approximately 6.8 billion yuan, already approaching the billion-dollar unicorn threshold.

As a transitional arrangement, Wangsu Science & Technology first provided 300 million yuan in convertible bonds to Sand.ai's domestic entity Shanghai Sandai Technology, and obtained A+ round warrants. After overseas investment (ODI) procedures are completed, it will exercise the rights to convert to shares. In other words, the 300 million yuan exists in debt form before delivery, and the company enjoys rights equivalent to preferred shares, leaving room on whether to convert. The announcement also clarified that the transaction does not constitute a related-party transaction or a major asset restructuring.

Small amount, low percentage, no consolidation—these are the three keywords of this deal. As of the end of June 2026, Wangsu Science & Technology had approximately 6.8 billion yuan in net cash on its books, with 300 million yuan accounting for less than 5% of the total. The security and value-added services segment had a gross margin of 74.03% in the first half, with overall cash flow stable and an asset-liability ratio of only 11.80%. In other words, even if this investment ultimately goes to waste, it would barely impact Wangsu Science & Technology's balance sheet.

A premium AI short drama costs only 30,000 to 200,000 yuan to produce, while mass-produced AI short dramas are compressed to 3,000 to 5,000 yuan. One to three people can churn out one in one to three days, while an S-grade live-action short drama still costs between 1.5 million and 3 million yuan to produce—a gap of up to 300 times.

But beneath the carnival, the bubble is also bursting. In the first half of the year, 221,900 AI dramas were released, but only 1,055 surpassed 100 million views—a 0.47% hit rate. Over 94% of works couldn't even reach a million views. Revenue per ten thousand views dropped from 30 to 100 yuan last year to just 5 to 10 yuan now, nearly halved twice over.

More troublesome is compliance. On September 2, 2026, the National Radio and Television Administration specifically convened a symposium on intellectual property and personality rights protection for AI micro short dramas regarding "face merging" and "voice merging," requiring that any face or voice merging behavior must obtain explicit authorization from rights holders. The "Micro Short Drama Development Management Measures" effective September 1 require AI dramas to add提示 labels in prominent positions in each episode. People's Daily also weighed in with a sharp critique that "AI faces are getting tiresome," and audiences are starting to complain about the monotonous "AI face."

The engine behind all these short dramas points to the same thing: video generation large models. And Sand.ai, which Wangsu Science & Technology just invested 300 million yuan in, is exactly a player in this track whose valuation has surged to a billion dollars. Piecing these two seemingly mismatched narratives together is the real highlight of this investment.

A "Micro Enterprise" Favored by Capital

Sand.ai, chosen by Wangsu Science & Technology, is actually closer to ordinary people than one might imagine. It has a product called VidMuse, centered on "Music in, Video Out." Users upload an audio clip, and the system automatically completes rhythm analysis, storyboard planning, and video generation. Within just two months of launch, its annualized revenue exceeded ten million dollars, making it one of the fastest products in the video agent track to reach this threshold.

In the words of Cao Yue's team themselves, the goal is to match the commercial needs of top short dramas in character consistency and multi-shot narrative. More critical is the cost. According to East Money's Caifuhao estimates for MAGI Preview, 0.5 yuan can generate a 10-second 1080P video, meaning video generation has shifted from occasional use to a daily production tool that can be called upon frequently.

Yet this company that turned technology into a product ordinary people can use is, according to business registration information, a typical "micro enterprise." Its operating entity Shanghai Sandai Technology Co., Ltd. was established in January 2024, with registered capital of 1 million yuan and paid-in capital of 100,000 yuan. But this small micro enterprise backs onto a "top-tier" academic and industry résumé.

Founder and CEO Cao Yue graduated from Tsinghua University with both his undergraduate and doctoral degrees, and received the Tsinghua Special Scholarship and Microsoft Research Asia Fellowship. He is a winner of the ICCV Best Paper Award (Marr Prize), one of the co-first authors of Swin Transformer, a former principal researcher at Microsoft Research Asia, head of the Visual Model Research Center at the Beijing Academy of Artificial Intelligence, and co-founder of Lightyear Beyond.

Co-founder Zhang Zheng is also a core author of Swin Transformer, and the core team originates from Microsoft Research Asia and Alibaba DAMO Academy. Sand.ai takes a non-consensus technical path. While the industry mainstream bets on Diffusion models, Cao Yue insists on autoregressive video generation. The company's self-developed Magi-1 model has long maintained first place in Google DeepMind's Physics IQ physical plausibility evaluation, and it is one of the earliest teams to produce a model with "simultaneous audio and video output."

Capital endorsement is equally strong. In June 2026, Sand.ai announced it had completed two consecutive funding rounds totaling over 100 million dollars. Investors include Kuaishou co-founder Su Hua, Meituan co-founder Wang Huiwen's family office (Lollapalooza Capital), as well as Matrix Partners China, Source Code Capital, IDG Capital, Innovation Works, Baidu Ventures, Jiukun Investment, and Heyu Capital, among a host of top-tier institutions. Earlier, the company also received seed funding from Source Code Capital and A-round funding from Capital Today. A company less than three years old assembling such a lineup shows just how hot the video generation track is.

Main Business Growth Has Peaked, Wangsu Science & Technology Needs a New Story

Wangsu Science & Technology isn't making a move because it lacks money. On the contrary, it has too much money and too few stories. Founded in 2000 and listed on the ChiNext board in October 2009, Wangsu Science & Technology is one of the earliest CDN (Content Delivery Network) service providers and has long been regarded as the A-share CDN leader. But its growth curve in recent years has flattened.

In the first half of 2026, the company achieved revenue of 2.318 billion yuan, down 1.38% year-on-year, and net profit attributable to shareholders of 348 million yuan, down 6.61% year-on-year. The company explained that the decline was mainly due to changes in the consolidation scope from the sale of subsidiary equity in the same period last year. On a comparable basis, revenue grew 13.11% year-on-year. Even excluding the accounting basis factor, the growth of the CDN main business has shown fatigue.

By segment, CDN and edge computing revenue in the first half was 1.368 billion yuan, accounting for 59.03%, down 9.52% year-on-year. Security and value-added services revenue was 790 million yuan, accounting for 34.06%, up 22.10% year-on-year, with a gross margin of 74.03%—the only bright spot. IDC and liquid cooling revenue was 128 million yuan, accounting for 5.52%. Overall gross margin was 35.92%, a six-year high, but both revenue scale and profit are under pressure.

For the full year 2025, Wangsu Science & Technology had revenue of 4.661 billion yuan and net profit attributable to shareholders of 800 million yuan. The scale is stable but lacks an explosive point. As of the end of June 2026, the company's market value was approximately 32.5 billion yuan, with nearly 7 billion yuan in net cash on its books. For a veteran manufacturer earning 800 million yuan a year, sitting on nearly 7 billion yuan in cash, yet with sluggish main business growth, spending 300 million yuan to bind a cutting-edge AI ecosystem partner looks more like telling a new story to the capital markets than an all-in gamble.

Wangsu Science & Technology's positioning is also very clear. The announcement states that the company is committed to becoming a globally leading edge intelligence and security service provider, and this investment aims to strengthen cooperation opportunities with video large model companies. Logically, video generation is one of the AI scenarios that consumes the most computing power and network bandwidth today. Every AI short drama you watch, from generation and rendering to distribution, must pass through massive edge nodes and content networks. Wangsu Science & Technology's global edge nodes, CDN distribution network, and security protection capabilities are perfectly positioned to become the underlying support for video models.

But the other side of the story is the several risk warnings laid out in the announcement. The transaction agreement has not yet been signed, and whether all parties can reach consensus and successfully sign remains uncertain. This investment involves subscribing to overseas enterprise shares and must complete ODI procedures, with the timely completion of registration and filing also in doubt. Even if the agreement is finalized, there is a possibility that delivery cannot be smoothly completed due to unfulfilled preconditions. More critically, the target is still in the early stages of product development and commercialization, and its future development depends on technology iteration, market environment, and industry competition.

In other words, whether the 300 million yuan can truly land and whether Sand.ai can succeed are still unknown. A more realistic challenge comes from the track itself. Video generation is no longer a blue ocean, and its cash-burning intensity far exceeds outside imagination. In September this year, OpenAI's Sora officially exited. Its standalone app closed on April 26, and its API went offline on September 24. Daily operating costs were about 1 million dollars, while cumulative in-app purchase revenue was only about 2.1 million dollars—making it a negative example in the industry showing that "technological leadership is not a moat; a commercial closed loop is."

The vacated position has mainly been taken over by Chinese manufacturers. Weekly data cited in a Huaxin Securities September research report shows that as of the week ending September 7, the ByteDance system ranked first globally in AI video model call volume market share at 41.6%, with computing power consumption share exceeding eighty percent. Kling surpassed 100 million global users, with second-quarter revenue exceeding 850 million yuan, up over 200% year-on-year, totaling about 1.5 billion yuan in the first half. Kuaishou even plans to spin it off for independent financing at a valuation of about 20 billion dollars. Seedance, backed by Douyin and Jianying, had monthly revenue exceeding 1 billion yuan in May alone. Alibaba's Tongyi Wanxiang, MiniMax's Hailuo, and Shengshu Technology's Vidu have also long formed a dense competitive belt.

Sand.ai must prove itself in the cracks between giants like ByteDance, Kuaishou, Alibaba, Tencent, and MiniMax—a difficult task. For Wangsu Science & Technology, extending edge computing, CDN distribution, and security capabilities into video generation scenarios is logically coherent but faces conversion challenges. A minority equity investment does not equal business binding. Which infrastructure provider a model company ultimately chooses depends on cost, performance, and ecosystem—not a piece of investment agreement.

When main business growth has peaked, what 300 million yuan buys may not be the future, but an observation window—using the smallest chip to stand at the edge of the most crowded and cash-burning table in AI video, watching how the cards are dealt. What do you think of Wangsu Science & Technology's investment? Feel free to discuss in the comments section.